Calibration
LABBench2 → Vals MMLU-Pro
Vals MMLU-Pro is estimated from LABBench2 with a Michaelis–Menten + offset curve fitted on 6 models measured on both: y = 0.0000 + 2.0000·x / (1.02637 + x), R² = 0.55, cross-validated error 1.4 pp. It is used for 1 estimate.
| Estimated model | LABBench2 | Vals MMLU-Pro | Source |
|---|---|---|---|
| Gemini 4 Argon | 88.8% | 92.8% | estimated ± 1.4 pp, low confidence |